The following algorithms have been implemented into the Supervised ML Classification toolbox:
- Discriminant Analysis (Quadratic: Maximum Likelihood)
- Naive Bayes
- Nearest Neighbors
- Random Forest
- Classification Trees
- Discriminant Analysis [ECOC]
- Kernel classification [ECOC]
- k-Nearest Neighbors [ECOC]
- Linear classification [ECOC]
- Naive Bayes Classification [ECOC]
- Decision Tree [ECOC]
- Support Vector Machines [ECOC]
- Discriminant Ensemble Learning
- KNN Ensemble Learning
- Tree Ensemble Learning [Bag]
- Tree Ensemble Learning [AdaBoostM1]
- Tree Ensemble Learning [RUSBoost]
- Neural Network [Adam]
- Pattern Recognition Network [trainlm]
- Extreme Learning Machine
- Pattern Recognition Network [trainbr]
- Pattern Recognition Network [trainscg]
- Gaussian Process Classification
In this first version, the classifiers can be combined with dimensionality reduction algorithms (e.g. PCA, PLS,...) and cross-validation options. It works for ENVI or (geo)TIFF images. The toolbox needs labeled data (labels & spectra), e.g. as prepared by the "LabelMeClass" tool.
A manual is still to be written. See the MLRA toolbox manual on how to use this toolbox, or contact us for assistance.
This toolbox has not been published yet. We will support if someone takes the initiative to explore and publish the toolbox.
Also, we look for people who would be interested to further develop this toolbox.